changepoint-detector had been checked on simulated counts and on real quarterly GDP, but not on a real count series with a documented break. The UK front-seat belt law took effect on 31 January 1983 (Harvey and Durbin 1986), and R's UKDriverDeaths holds 192 monthly totals of drivers killed or seriously injured, 1969 to 1984.
With seasonal negative-binomial counts the detector returns 8 changes. One is 1983-01, from about 1600 a month to 1329, with a 95% range of 1983-01 to 1983-02, which holds the first full month under the law. Another is the winter-1973 dip. Plain Poisson on the same data returns 33 changes, because the months are clumpier than a Poisson allows. That is the case for --overdispersion nb.
One series and one break shows the range is plausible on real data. It does not show calibration; the seeded simulations do that. Script, test and data source are in the venture's eval/ and data/SOURCE.md. Statistical analysis, not advice.